Systems that stay up.▌

Built an AWS AIOps platform in Django covering 8+ incident types — it ingests alarms, runs EC2 diagnostics, writes the RCA, and auto-remediates approved CPU, memory, disk-log and service-health issues on its own.

Engineered real-time bad-bot detection on live mobile-site traffic: Varnish/Fluentd logs aggregated in 10-minute Redis windows, judged by rules + Isolation Forest + fingerprinting + crawler verification.

Automated monitoring across a 5-node, ~45TB Elastic/Kibana stack — Query DSL + pandas pushing alerts to Sheets, Chat and email the moment 5xx, latency or traffic misbehaves.

Architected the multilingual voice-bot backend for IPV and Anarock — Django, SIP, ElevenLabs, OpenAI, Deepgram Nova-3 and Llama analysis, conversing in Hindi, English, Kannada and Marathi.

Scaled outbound to 30 concurrent calls and 10,000+ leads/day per client, with summaries, intent, sentiment and lead analysis landing in an in-house CRM (external CRMs supported).

Pushed real-time call-assist from 10–12 to 40+ live calls by rebuilding Redis/Celery queues, WebSocket streams, pub/sub partitioning, async audio, caching and backpressure.

Cut nudge latency from 2.8s to under 900ms and Redis load by ~60% — live transcription, agent nudges and post-call automation stopped wobbling.

Shipped a GMeet interview bot — Puppeteer, transcription, diarization, Llama scoring — on AWS load-balanced infrastructure. HR review effort dropped ~50%.

Designed a retail audio-intelligence pipeline (FastAPI, Celery, Docker) that chews through 24 hours of store audio every day — batch speed up 30%, transcription quality up ~20%.